volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #define _XOPEN_SOURCE 600
  19 #define __USE_XOPEN2K
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <uuid/uuid.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "print-tree.h"
  31 #include "volumes.h"
  32 #include "math.h"
  33
  34 struct stripe {
  35         struct btrfs_device *dev;
  36         u64 physical;
  37 };
  38
  39 static inline int nr_parity_stripes(struct map_lookup *map)
  40 {
  41         if (map->type & BTRFS_BLOCK_GROUP_RAID5)
  42                 return 1;
  43         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
  44                 return 2;
  45         else
  46                 return 0;
  47 }
  48
  49 static inline int nr_data_stripes(struct map_lookup *map)
  50 {
  51         return map->num_stripes - nr_parity_stripes(map);
  52 }
  53
  54 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
  55
  56 static LIST_HEAD(fs_uuids);
  57
  58 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  59                                           u8 *uuid)
  60 {
  61         struct btrfs_device *dev;
  62         struct list_head *cur;
  63
  64         list_for_each(cur, head) {
  65                 dev = list_entry(cur, struct btrfs_device, dev_list);
  66                 if (dev->devid == devid &&
  67                     !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
  68                         return dev;
  69                 }
  70         }
  71         return NULL;
  72 }
  73
  74 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  75 {
  76         struct list_head *cur;
  77         struct btrfs_fs_devices *fs_devices;
  78
  79         list_for_each(cur, &fs_uuids) {
  80                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  81                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  82                         return fs_devices;
  83         }
  84         return NULL;
  85 }
  86
  87 static int device_list_add(const char *path,
  88                            struct btrfs_super_block *disk_super,
  89                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
  90 {
  91         struct btrfs_device *device;
  92         struct btrfs_fs_devices *fs_devices;
  93         u64 found_transid = btrfs_super_generation(disk_super);
  94
  95         fs_devices = find_fsid(disk_super->fsid);
  96         if (!fs_devices) {
  97                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
  98                 if (!fs_devices)
  99                         return -ENOMEM;
 100                 INIT_LIST_HEAD(&fs_devices->devices);
 101                 list_add(&fs_devices->list, &fs_uuids);
 102                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 103                 fs_devices->latest_devid = devid;
 104                 fs_devices->latest_trans = found_transid;
 105                 fs_devices->lowest_devid = (u64)-1;
 106                 device = NULL;
 107         } else {
 108                 device = __find_device(&fs_devices->devices, devid,
 109                                        disk_super->dev_item.uuid);
 110         }
 111         if (!device) {
 112                 device = kzalloc(sizeof(*device), GFP_NOFS);
 113                 if (!device) {
 114                         /* we can safely leave the fs_devices entry around */
 115                         return -ENOMEM;
 116                 }
 117                 device->fd = -1;
 118                 device->devid = devid;
 119                 memcpy(device->uuid, disk_super->dev_item.uuid,
 120                        BTRFS_UUID_SIZE);
 121                 device->name = kstrdup(path, GFP_NOFS);
 122                 if (!device->name) {
 123                         kfree(device);
 124                         return -ENOMEM;
 125                 }
 126                 device->label = kstrdup(disk_super->label, GFP_NOFS);
 127                 device->total_devs = btrfs_super_num_devices(disk_super);
 128                 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
 129                 device->total_bytes =
 130                         btrfs_stack_device_total_bytes(&disk_super->dev_item);
 131                 device->bytes_used =
 132                         btrfs_stack_device_bytes_used(&disk_super->dev_item);
 133                 list_add(&device->dev_list, &fs_devices->devices);
 134                 device->fs_devices = fs_devices;
 135         } else if (!device->name || strcmp(device->name, path)) {
 136                 char *name = strdup(path);
 137                 if (!name)
 138                         return -ENOMEM;
 139                 kfree(device->name);
 140                 device->name = name;
 141         }
 142
 143
 144         if (found_transid > fs_devices->latest_trans) {
 145                 fs_devices->latest_devid = devid;
 146                 fs_devices->latest_trans = found_transid;
 147         }
 148         if (fs_devices->lowest_devid > devid) {
 149                 fs_devices->lowest_devid = devid;
 150         }
 151         *fs_devices_ret = fs_devices;
 152         return 0;
 153 }
 154
 155 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 156 {
 157         struct btrfs_fs_devices *seed_devices;
 158         struct list_head *cur;
 159         struct btrfs_device *device;
 160 again:
 161         list_for_each(cur, &fs_devices->devices) {
 162                 device = list_entry(cur, struct btrfs_device, dev_list);
 163                 if (device->fd != -1) {
 164                         fsync(device->fd);
 165                         if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
 166                                 fprintf(stderr, "Warning, could not drop caches\n");
 167                         close(device->fd);
 168                         device->fd = -1;
 169                 }
 170                 device->writeable = 0;
 171         }
 172
 173         seed_devices = fs_devices->seed;
 174         fs_devices->seed = NULL;
 175         if (seed_devices) {
 176                 fs_devices = seed_devices;
 177                 goto again;
 178         }
 179
 180         return 0;
 181 }
 182
 183 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 184 {
 185         int fd;
 186         struct list_head *head = &fs_devices->devices;
 187         struct list_head *cur;
 188         struct btrfs_device *device;
 189         int ret;
 190
 191         list_for_each(cur, head) {
 192                 device = list_entry(cur, struct btrfs_device, dev_list);
 193                 if (!device->name) {
 194                         printk("no name for device %llu, skip it now\n", device->devid);
 195                         continue;
 196                 }
 197
 198                 fd = open(device->name, flags);
 199                 if (fd < 0) {
 200                         ret = -errno;
 201                         goto fail;
 202                 }
 203
 204                 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
 205                         fprintf(stderr, "Warning, could not drop caches\n");
 206
 207                 if (device->devid == fs_devices->latest_devid)
 208                         fs_devices->latest_bdev = fd;
 209                 if (device->devid == fs_devices->lowest_devid)
 210                         fs_devices->lowest_bdev = fd;
 211                 device->fd = fd;
 212                 if (flags == O_RDWR)
 213                         device->writeable = 1;
 214         }
 215         return 0;
 216 fail:
 217         btrfs_close_devices(fs_devices);
 218         return ret;
 219 }
 220
 221 int btrfs_scan_one_device(int fd, const char *path,
 222                           struct btrfs_fs_devices **fs_devices_ret,
 223                           u64 *total_devs, u64 super_offset)
 224 {
 225         struct btrfs_super_block *disk_super;
 226         char *buf;
 227         int ret;
 228         u64 devid;
 229         char uuidbuf[37];
 230
 231         buf = malloc(4096);
 232         if (!buf) {
 233                 ret = -ENOMEM;
 234                 goto error;
 235         }
 236         disk_super = (struct btrfs_super_block *)buf;
 237         ret = btrfs_read_dev_super(fd, disk_super, super_offset);
 238         if (ret < 0) {
 239                 ret = -EIO;
 240                 goto error_brelse;
 241         }
 242         devid = btrfs_stack_device_id(&disk_super->dev_item);
 243         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
 244                 *total_devs = 1;
 245         else
 246                 *total_devs = btrfs_super_num_devices(disk_super);
 247         uuid_unparse(disk_super->fsid, uuidbuf);
 248
 249         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 250
 251 error_brelse:
 252         free(buf);
 253 error:
 254         return ret;
 255 }
 256
 257 /*
 258  * this uses a pretty simple search, the expectation is that it is
 259  * called very infrequently and that a given device has a small number
 260  * of extents
 261  */
 262 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 263                                 struct btrfs_device *device,
 264                                 struct btrfs_path *path,
 265                                 u64 num_bytes, u64 *start)
 266 {
 267         struct btrfs_key key;
 268         struct btrfs_root *root = device->dev_root;
 269         struct btrfs_dev_extent *dev_extent = NULL;
 270         u64 hole_size = 0;
 271         u64 last_byte = 0;
 272         u64 search_start = 0;
 273         u64 search_end = device->total_bytes;
 274         int ret;
 275         int slot = 0;
 276         int start_found;
 277         struct extent_buffer *l;
 278
 279         start_found = 0;
 280         path->reada = 2;
 281
 282         /* FIXME use last free of some kind */
 283
 284         /* we don't want to overwrite the superblock on the drive,
 285          * so we make sure to start at an offset of at least 1MB
 286          */
 287         search_start = max((u64)1024 * 1024, search_start);
 288
 289         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 290                 search_start = max(root->fs_info->alloc_start, search_start);
 291
 292         key.objectid = device->devid;
 293         key.offset = search_start;
 294         key.type = BTRFS_DEV_EXTENT_KEY;
 295         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 296         if (ret < 0)
 297                 goto error;
 298         ret = btrfs_previous_item(root, path, 0, key.type);
 299         if (ret < 0)
 300                 goto error;
 301         l = path->nodes[0];
 302         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 303         while (1) {
 304                 l = path->nodes[0];
 305                 slot = path->slots[0];
 306                 if (slot >= btrfs_header_nritems(l)) {
 307                         ret = btrfs_next_leaf(root, path);
 308                         if (ret == 0)
 309                                 continue;
 310                         if (ret < 0)
 311                                 goto error;
 312 no_more_items:
 313                         if (!start_found) {
 314                                 if (search_start >= search_end) {
 315                                         ret = -ENOSPC;
 316                                         goto error;
 317                                 }
 318                                 *start = search_start;
 319                                 start_found = 1;
 320                                 goto check_pending;
 321                         }
 322                         *start = last_byte > search_start ?
 323                                 last_byte : search_start;
 324                         if (search_end <= *start) {
 325                                 ret = -ENOSPC;
 326                                 goto error;
 327                         }
 328                         goto check_pending;
 329                 }
 330                 btrfs_item_key_to_cpu(l, &key, slot);
 331
 332                 if (key.objectid < device->devid)
 333                         goto next;
 334
 335                 if (key.objectid > device->devid)
 336                         goto no_more_items;
 337
 338                 if (key.offset >= search_start && key.offset > last_byte &&
 339                     start_found) {
 340                         if (last_byte < search_start)
 341                                 last_byte = search_start;
 342                         hole_size = key.offset - last_byte;
 343                         if (key.offset > last_byte &&
 344                             hole_size >= num_bytes) {
 345                                 *start = last_byte;
 346                                 goto check_pending;
 347                         }
 348                 }
 349                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 350                         goto next;
 351                 }
 352
 353                 start_found = 1;
 354                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 355                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 356 next:
 357                 path->slots[0]++;
 358                 cond_resched();
 359         }
 360 check_pending:
 361         /* we have to make sure we didn't find an extent that has already
 362          * been allocated by the map tree or the original allocation
 363          */
 364         btrfs_release_path(root, path);
 365         BUG_ON(*start < search_start);
 366
 367         if (*start + num_bytes > search_end) {
 368                 ret = -ENOSPC;
 369                 goto error;
 370         }
 371         /* check for pending inserts here */
 372         return 0;
 373
 374 error:
 375         btrfs_release_path(root, path);
 376         return ret;
 377 }
 378
 379 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 380                            struct btrfs_device *device,
 381                            u64 chunk_tree, u64 chunk_objectid,
 382                            u64 chunk_offset,
 383                            u64 num_bytes, u64 *start)
 384 {
 385         int ret;
 386         struct btrfs_path *path;
 387         struct btrfs_root *root = device->dev_root;
 388         struct btrfs_dev_extent *extent;
 389         struct extent_buffer *leaf;
 390         struct btrfs_key key;
 391
 392         path = btrfs_alloc_path();
 393         if (!path)
 394                 return -ENOMEM;
 395
 396         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 397         if (ret) {
 398                 goto err;
 399         }
 400
 401         key.objectid = device->devid;
 402         key.offset = *start;
 403         key.type = BTRFS_DEV_EXTENT_KEY;
 404         ret = btrfs_insert_empty_item(trans, root, path, &key,
 405                                       sizeof(*extent));
 406         BUG_ON(ret);
 407
 408         leaf = path->nodes[0];
 409         extent = btrfs_item_ptr(leaf, path->slots[0],
 410                                 struct btrfs_dev_extent);
 411         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 412         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 413         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 414
 415         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 416                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 417                     BTRFS_UUID_SIZE);
 418
 419         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 420         btrfs_mark_buffer_dirty(leaf);
 421 err:
 422         btrfs_free_path(path);
 423         return ret;
 424 }
 425
 426 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 427 {
 428         struct btrfs_path *path;
 429         int ret;
 430         struct btrfs_key key;
 431         struct btrfs_chunk *chunk;
 432         struct btrfs_key found_key;
 433
 434         path = btrfs_alloc_path();
 435         BUG_ON(!path);
 436
 437         key.objectid = objectid;
 438         key.offset = (u64)-1;
 439         key.type = BTRFS_CHUNK_ITEM_KEY;
 440
 441         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 442         if (ret < 0)
 443                 goto error;
 444
 445         BUG_ON(ret == 0);
 446
 447         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 448         if (ret) {
 449                 *offset = 0;
 450         } else {
 451                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 452                                       path->slots[0]);
 453                 if (found_key.objectid != objectid)
 454                         *offset = 0;
 455                 else {
 456                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 457                                                struct btrfs_chunk);
 458                         *offset = found_key.offset +
 459                                 btrfs_chunk_length(path->nodes[0], chunk);
 460                 }
 461         }
 462         ret = 0;
 463 error:
 464         btrfs_free_path(path);
 465         return ret;
 466 }
 467
 468 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 469                            u64 *objectid)
 470 {
 471         int ret;
 472         struct btrfs_key key;
 473         struct btrfs_key found_key;
 474
 475         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 476         key.type = BTRFS_DEV_ITEM_KEY;
 477         key.offset = (u64)-1;
 478
 479         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 480         if (ret < 0)
 481                 goto error;
 482
 483         BUG_ON(ret == 0);
 484
 485         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 486                                   BTRFS_DEV_ITEM_KEY);
 487         if (ret) {
 488                 *objectid = 1;
 489         } else {
 490                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 491                                       path->slots[0]);
 492                 *objectid = found_key.offset + 1;
 493         }
 494         ret = 0;
 495 error:
 496         btrfs_release_path(root, path);
 497         return ret;
 498 }
 499
 500 /*
 501  * the device information is stored in the chunk root
 502  * the btrfs_device struct should be fully filled in
 503  */
 504 int btrfs_add_device(struct btrfs_trans_handle *trans,
 505                      struct btrfs_root *root,
 506                      struct btrfs_device *device)
 507 {
 508         int ret;
 509         struct btrfs_path *path;
 510         struct btrfs_dev_item *dev_item;
 511         struct extent_buffer *leaf;
 512         struct btrfs_key key;
 513         unsigned long ptr;
 514         u64 free_devid = 0;
 515
 516         root = root->fs_info->chunk_root;
 517
 518         path = btrfs_alloc_path();
 519         if (!path)
 520                 return -ENOMEM;
 521
 522         ret = find_next_devid(root, path, &free_devid);
 523         if (ret)
 524                 goto out;
 525
 526         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 527         key.type = BTRFS_DEV_ITEM_KEY;
 528         key.offset = free_devid;
 529
 530         ret = btrfs_insert_empty_item(trans, root, path, &key,
 531                                       sizeof(*dev_item));
 532         if (ret)
 533                 goto out;
 534
 535         leaf = path->nodes[0];
 536         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 537
 538         device->devid = free_devid;
 539         btrfs_set_device_id(leaf, dev_item, device->devid);
 540         btrfs_set_device_generation(leaf, dev_item, 0);
 541         btrfs_set_device_type(leaf, dev_item, device->type);
 542         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 543         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 544         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 545         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 546         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 547         btrfs_set_device_group(leaf, dev_item, 0);
 548         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 549         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 550         btrfs_set_device_start_offset(leaf, dev_item, 0);
 551
 552         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 553         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 554         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 555         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 556         btrfs_mark_buffer_dirty(leaf);
 557         ret = 0;
 558
 559 out:
 560         btrfs_free_path(path);
 561         return ret;
 562 }
 563
 564 int btrfs_update_device(struct btrfs_trans_handle *trans,
 565                         struct btrfs_device *device)
 566 {
 567         int ret;
 568         struct btrfs_path *path;
 569         struct btrfs_root *root;
 570         struct btrfs_dev_item *dev_item;
 571         struct extent_buffer *leaf;
 572         struct btrfs_key key;
 573
 574         root = device->dev_root->fs_info->chunk_root;
 575
 576         path = btrfs_alloc_path();
 577         if (!path)
 578                 return -ENOMEM;
 579
 580         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 581         key.type = BTRFS_DEV_ITEM_KEY;
 582         key.offset = device->devid;
 583
 584         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 585         if (ret < 0)
 586                 goto out;
 587
 588         if (ret > 0) {
 589                 ret = -ENOENT;
 590                 goto out;
 591         }
 592
 593         leaf = path->nodes[0];
 594         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 595
 596         btrfs_set_device_id(leaf, dev_item, device->devid);
 597         btrfs_set_device_type(leaf, dev_item, device->type);
 598         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 599         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 600         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 601         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 602         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 603         btrfs_mark_buffer_dirty(leaf);
 604
 605 out:
 606         btrfs_free_path(path);
 607         return ret;
 608 }
 609
 610 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 611                            struct btrfs_root *root,
 612                            struct btrfs_key *key,
 613                            struct btrfs_chunk *chunk, int item_size)
 614 {
 615         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
 616         struct btrfs_disk_key disk_key;
 617         u32 array_size;
 618         u8 *ptr;
 619
 620         array_size = btrfs_super_sys_array_size(super_copy);
 621         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 622                 return -EFBIG;
 623
 624         ptr = super_copy->sys_chunk_array + array_size;
 625         btrfs_cpu_key_to_disk(&disk_key, key);
 626         memcpy(ptr, &disk_key, sizeof(disk_key));
 627         ptr += sizeof(disk_key);
 628         memcpy(ptr, chunk, item_size);
 629         item_size += sizeof(disk_key);
 630         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 631         return 0;
 632 }
 633
 634 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 635                                int sub_stripes)
 636 {
 637         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 638                 return calc_size;
 639         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 640                 return calc_size * (num_stripes / sub_stripes);
 641         else if (type & BTRFS_BLOCK_GROUP_RAID5)
 642                 return calc_size * (num_stripes - 1);
 643         else if (type & BTRFS_BLOCK_GROUP_RAID6)
 644                 return calc_size * (num_stripes - 2);
 645         else
 646                 return calc_size * num_stripes;
 647 }
 648
 649
 650 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
 651 {
 652         /* TODO, add a way to store the preferred stripe size */
 653         return 64 * 1024;
 654 }
 655
 656 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 657                       struct btrfs_root *extent_root, u64 *start,
 658                       u64 *num_bytes, u64 type)
 659 {
 660         u64 dev_offset;
 661         struct btrfs_fs_info *info = extent_root->fs_info;
 662         struct btrfs_root *chunk_root = info->chunk_root;
 663         struct btrfs_stripe *stripes;
 664         struct btrfs_device *device = NULL;
 665         struct btrfs_chunk *chunk;
 666         struct list_head private_devs;
 667         struct list_head *dev_list = &info->fs_devices->devices;
 668         struct list_head *cur;
 669         struct map_lookup *map;
 670         int min_stripe_size = 1 * 1024 * 1024;
 671         u64 calc_size = 8 * 1024 * 1024;
 672         u64 min_free;
 673         u64 max_chunk_size = 4 * calc_size;
 674         u64 avail;
 675         u64 max_avail = 0;
 676         u64 percent_max;
 677         int num_stripes = 1;
 678         int min_stripes = 1;
 679         int sub_stripes = 0;
 680         int looped = 0;
 681         int ret;
 682         int index;
 683         int stripe_len = 64 * 1024;
 684         struct btrfs_key key;
 685         u64 offset;
 686
 687         if (list_empty(dev_list)) {
 688                 return -ENOSPC;
 689         }
 690
 691         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 692                     BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
 693                     BTRFS_BLOCK_GROUP_RAID10 |
 694                     BTRFS_BLOCK_GROUP_DUP)) {
 695                 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 696                         calc_size = 8 * 1024 * 1024;
 697                         max_chunk_size = calc_size * 2;
 698                         min_stripe_size = 1 * 1024 * 1024;
 699                 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
 700                         calc_size = 1024 * 1024 * 1024;
 701                         max_chunk_size = 10 * calc_size;
 702                         min_stripe_size = 64 * 1024 * 1024;
 703                 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 704                         calc_size = 1024 * 1024 * 1024;
 705                         max_chunk_size = 4 * calc_size;
 706                         min_stripe_size = 32 * 1024 * 1024;
 707                 }
 708         }
 709         if (type & BTRFS_BLOCK_GROUP_RAID1) {
 710                 num_stripes = min_t(u64, 2,
 711                                   btrfs_super_num_devices(info->super_copy));
 712                 if (num_stripes < 2)
 713                         return -ENOSPC;
 714                 min_stripes = 2;
 715         }
 716         if (type & BTRFS_BLOCK_GROUP_DUP) {
 717                 num_stripes = 2;
 718                 min_stripes = 2;
 719         }
 720         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 721                 num_stripes = btrfs_super_num_devices(info->super_copy);
 722                 min_stripes = 2;
 723         }
 724         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 725                 num_stripes = btrfs_super_num_devices(info->super_copy);
 726                 if (num_stripes < 4)
 727                         return -ENOSPC;
 728                 num_stripes &= ~(u32)1;
 729                 sub_stripes = 2;
 730                 min_stripes = 4;
 731         }
 732         if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
 733                 num_stripes = btrfs_super_num_devices(info->super_copy);
 734                 if (num_stripes < 2)
 735                         return -ENOSPC;
 736                 min_stripes = 2;
 737                 stripe_len = find_raid56_stripe_len(num_stripes - 1,
 738                                     btrfs_super_stripesize(info->super_copy));
 739         }
 740         if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
 741                 num_stripes = btrfs_super_num_devices(info->super_copy);
 742                 if (num_stripes < 3)
 743                         return -ENOSPC;
 744                 min_stripes = 3;
 745                 stripe_len = find_raid56_stripe_len(num_stripes - 2,
 746                                     btrfs_super_stripesize(info->super_copy));
 747         }
 748
 749         /* we don't want a chunk larger than 10% of the FS */
 750         percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
 751         max_chunk_size = min(percent_max, max_chunk_size);
 752
 753 again:
 754         if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
 755             max_chunk_size) {
 756                 calc_size = max_chunk_size;
 757                 calc_size /= num_stripes;
 758                 calc_size /= stripe_len;
 759                 calc_size *= stripe_len;
 760         }
 761         /* we don't want tiny stripes */
 762         calc_size = max_t(u64, calc_size, min_stripe_size);
 763
 764         calc_size /= stripe_len;
 765         calc_size *= stripe_len;
 766         INIT_LIST_HEAD(&private_devs);
 767         cur = dev_list->next;
 768         index = 0;
 769
 770         if (type & BTRFS_BLOCK_GROUP_DUP)
 771                 min_free = calc_size * 2;
 772         else
 773                 min_free = calc_size;
 774
 775         /* build a private list of devices we will allocate from */
 776         while(index < num_stripes) {
 777                 device = list_entry(cur, struct btrfs_device, dev_list);
 778                 avail = device->total_bytes - device->bytes_used;
 779                 cur = cur->next;
 780                 if (avail >= min_free) {
 781                         list_move_tail(&device->dev_list, &private_devs);
 782                         index++;
 783                         if (type & BTRFS_BLOCK_GROUP_DUP)
 784                                 index++;
 785                 } else if (avail > max_avail)
 786                         max_avail = avail;
 787                 if (cur == dev_list)
 788                         break;
 789         }
 790         if (index < num_stripes) {
 791                 list_splice(&private_devs, dev_list);
 792                 if (index >= min_stripes) {
 793                         num_stripes = index;
 794                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 795                                 num_stripes /= sub_stripes;
 796                                 num_stripes *= sub_stripes;
 797                         }
 798                         looped = 1;
 799                         goto again;
 800                 }
 801                 if (!looped && max_avail > 0) {
 802                         looped = 1;
 803                         calc_size = max_avail;
 804                         goto again;
 805                 }
 806                 return -ENOSPC;
 807         }
 808         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 809                               &offset);
 810         if (ret)
 811                 return ret;
 812         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 813         key.type = BTRFS_CHUNK_ITEM_KEY;
 814         key.offset = offset;
 815
 816         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 817         if (!chunk)
 818                 return -ENOMEM;
 819
 820         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
 821         if (!map) {
 822                 kfree(chunk);
 823                 return -ENOMEM;
 824         }
 825
 826         stripes = &chunk->stripe;
 827         *num_bytes = chunk_bytes_by_type(type, calc_size,
 828                                          num_stripes, sub_stripes);
 829         index = 0;
 830         while(index < num_stripes) {
 831                 struct btrfs_stripe *stripe;
 832                 BUG_ON(list_empty(&private_devs));
 833                 cur = private_devs.next;
 834                 device = list_entry(cur, struct btrfs_device, dev_list);
 835
 836                 /* loop over this device again if we're doing a dup group */
 837                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 838                     (index == num_stripes - 1))
 839                         list_move_tail(&device->dev_list, dev_list);
 840
 841                 ret = btrfs_alloc_dev_extent(trans, device,
 842                              info->chunk_root->root_key.objectid,
 843                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 844                              calc_size, &dev_offset);
 845                 BUG_ON(ret);
 846
 847                 device->bytes_used += calc_size;
 848                 ret = btrfs_update_device(trans, device);
 849                 BUG_ON(ret);
 850
 851                 map->stripes[index].dev = device;
 852                 map->stripes[index].physical = dev_offset;
 853                 stripe = stripes + index;
 854                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 855                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 856                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 857                 index++;
 858         }
 859         BUG_ON(!list_empty(&private_devs));
 860
 861         /* key was set above */
 862         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 863         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 864         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 865         btrfs_set_stack_chunk_type(chunk, type);
 866         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 867         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 868         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 869         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 870         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 871         map->sector_size = extent_root->sectorsize;
 872         map->stripe_len = stripe_len;
 873         map->io_align = stripe_len;
 874         map->io_width = stripe_len;
 875         map->type = type;
 876         map->num_stripes = num_stripes;
 877         map->sub_stripes = sub_stripes;
 878
 879         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 880                                 btrfs_chunk_item_size(num_stripes));
 881         BUG_ON(ret);
 882         *start = key.offset;;
 883
 884         map->ce.start = key.offset;
 885         map->ce.size = *num_bytes;
 886
 887         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
 888         BUG_ON(ret);
 889
 890         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 891                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
 892                                     chunk, btrfs_chunk_item_size(num_stripes));
 893                 BUG_ON(ret);
 894         }
 895
 896         kfree(chunk);
 897         return ret;
 898 }
 899
 900 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
 901                            struct btrfs_root *extent_root, u64 *start,
 902                            u64 num_bytes, u64 type)
 903 {
 904         u64 dev_offset;
 905         struct btrfs_fs_info *info = extent_root->fs_info;
 906         struct btrfs_root *chunk_root = info->chunk_root;
 907         struct btrfs_stripe *stripes;
 908         struct btrfs_device *device = NULL;
 909         struct btrfs_chunk *chunk;
 910         struct list_head *dev_list = &info->fs_devices->devices;
 911         struct list_head *cur;
 912         struct map_lookup *map;
 913         u64 calc_size = 8 * 1024 * 1024;
 914         int num_stripes = 1;
 915         int sub_stripes = 0;
 916         int ret;
 917         int index;
 918         int stripe_len = 64 * 1024;
 919         struct btrfs_key key;
 920
 921         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 922         key.type = BTRFS_CHUNK_ITEM_KEY;
 923         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 924                               &key.offset);
 925         if (ret)
 926                 return ret;
 927
 928         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 929         if (!chunk)
 930                 return -ENOMEM;
 931
 932         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
 933         if (!map) {
 934                 kfree(chunk);
 935                 return -ENOMEM;
 936         }
 937
 938         stripes = &chunk->stripe;
 939         calc_size = num_bytes;
 940
 941         index = 0;
 942         cur = dev_list->next;
 943         device = list_entry(cur, struct btrfs_device, dev_list);
 944
 945         while (index < num_stripes) {
 946                 struct btrfs_stripe *stripe;
 947
 948                 ret = btrfs_alloc_dev_extent(trans, device,
 949                              info->chunk_root->root_key.objectid,
 950                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 951                              calc_size, &dev_offset);
 952                 BUG_ON(ret);
 953
 954                 device->bytes_used += calc_size;
 955                 ret = btrfs_update_device(trans, device);
 956                 BUG_ON(ret);
 957
 958                 map->stripes[index].dev = device;
 959                 map->stripes[index].physical = dev_offset;
 960                 stripe = stripes + index;
 961                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 962                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 963                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 964                 index++;
 965         }
 966
 967         /* key was set above */
 968         btrfs_set_stack_chunk_length(chunk, num_bytes);
 969         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 970         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 971         btrfs_set_stack_chunk_type(chunk, type);
 972         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 973         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 974         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 975         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 976         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 977         map->sector_size = extent_root->sectorsize;
 978         map->stripe_len = stripe_len;
 979         map->io_align = stripe_len;
 980         map->io_width = stripe_len;
 981         map->type = type;
 982         map->num_stripes = num_stripes;
 983         map->sub_stripes = sub_stripes;
 984
 985         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 986                                 btrfs_chunk_item_size(num_stripes));
 987         BUG_ON(ret);
 988         *start = key.offset;
 989
 990         map->ce.start = key.offset;
 991         map->ce.size = num_bytes;
 992
 993         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
 994         BUG_ON(ret);
 995
 996         kfree(chunk);
 997         return ret;
 998 }
 999
1000 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1001 {
1002         cache_tree_init(&tree->cache_tree);
1003 }
1004
1005 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1006 {
1007         struct cache_extent *ce;
1008         struct map_lookup *map;
1009         int ret;
1010
1011         ce = search_cache_extent(&map_tree->cache_tree, logical);
1012         BUG_ON(!ce);
1013         BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1014         map = container_of(ce, struct map_lookup, ce);
1015
1016         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1017                 ret = map->num_stripes;
1018         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1019                 ret = map->sub_stripes;
1020         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1021                 ret = 2;
1022         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1023                 ret = 3;
1024         else
1025                 ret = 1;
1026         return ret;
1027 }
1028
1029 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1030                         u64 *size)
1031 {
1032         struct cache_extent *ce;
1033         struct map_lookup *map;
1034
1035         ce = search_cache_extent(&map_tree->cache_tree, *logical);
1036
1037         while (ce) {
1038                 ce = next_cache_extent(ce);
1039                 if (!ce)
1040                         return -ENOENT;
1041
1042                 map = container_of(ce, struct map_lookup, ce);
1043                 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1044                         *logical = ce->start;
1045                         *size = ce->size;
1046                         return 0;
1047                 }
1048         }
1049
1050         return -ENOENT;
1051 }
1052
1053 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1054                      u64 chunk_start, u64 physical, u64 devid,
1055                      u64 **logical, int *naddrs, int *stripe_len)
1056 {
1057         struct cache_extent *ce;
1058         struct map_lookup *map;
1059         u64 *buf;
1060         u64 bytenr;
1061         u64 length;
1062         u64 stripe_nr;
1063         u64 rmap_len;
1064         int i, j, nr = 0;
1065
1066         ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1067         BUG_ON(!ce);
1068         map = container_of(ce, struct map_lookup, ce);
1069
1070         length = ce->size;
1071         rmap_len = map->stripe_len;
1072         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1073                 length = ce->size / (map->num_stripes / map->sub_stripes);
1074         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1075                 length = ce->size / map->num_stripes;
1076         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1077                               BTRFS_BLOCK_GROUP_RAID6)) {
1078                 length = ce->size / nr_data_stripes(map);
1079                 rmap_len = map->stripe_len * nr_data_stripes(map);
1080         }
1081
1082         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1083
1084         for (i = 0; i < map->num_stripes; i++) {
1085                 if (devid && map->stripes[i].dev->devid != devid)
1086                         continue;
1087                 if (map->stripes[i].physical > physical ||
1088                     map->stripes[i].physical + length <= physical)
1089                         continue;
1090
1091                 stripe_nr = (physical - map->stripes[i].physical) /
1092                             map->stripe_len;
1093
1094                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1095                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1096                                     map->sub_stripes;
1097                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1098                         stripe_nr = stripe_nr * map->num_stripes + i;
1099                 } /* else if RAID[56], multiply by nr_data_stripes().
1100                    * Alternatively, just use rmap_len below instead of
1101                    * map->stripe_len */
1102
1103                 bytenr = ce->start + stripe_nr * rmap_len;
1104                 for (j = 0; j < nr; j++) {
1105                         if (buf[j] == bytenr)
1106                                 break;
1107                 }
1108                 if (j == nr)
1109                         buf[nr++] = bytenr;
1110         }
1111
1112         *logical = buf;
1113         *naddrs = nr;
1114         *stripe_len = rmap_len;
1115
1116         return 0;
1117 }
1118
1119 static inline int parity_smaller(u64 a, u64 b)
1120 {
1121         return a > b;
1122 }
1123
1124 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1125 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1126 {
1127         struct btrfs_bio_stripe s;
1128         int i;
1129         u64 l;
1130         int again = 1;
1131
1132         while (again) {
1133                 again = 0;
1134                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1135                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1136                                 s = bbio->stripes[i];
1137                                 l = raid_map[i];
1138                                 bbio->stripes[i] = bbio->stripes[i+1];
1139                                 raid_map[i] = raid_map[i+1];
1140                                 bbio->stripes[i+1] = s;
1141                                 raid_map[i+1] = l;
1142                                 again = 1;
1143                         }
1144                 }
1145         }
1146 }
1147
1148 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1149                     u64 logical, u64 *length,
1150                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1151                     u64 **raid_map_ret)
1152 {
1153         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1154                                  multi_ret, mirror_num, raid_map_ret);
1155 }
1156
1157 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1158                     u64 logical, u64 *length, u64 *type,
1159                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1160                     u64 **raid_map_ret)
1161 {
1162         struct cache_extent *ce;
1163         struct map_lookup *map;
1164         u64 offset;
1165         u64 stripe_offset;
1166         u64 stripe_nr;
1167         u64 *raid_map = NULL;
1168         int stripes_allocated = 8;
1169         int stripes_required = 1;
1170         int stripe_index;
1171         int i;
1172         struct btrfs_multi_bio *multi = NULL;
1173
1174         if (multi_ret && rw == READ) {
1175                 stripes_allocated = 1;
1176         }
1177 again:
1178         ce = search_cache_extent(&map_tree->cache_tree, logical);
1179         if (!ce) {
1180                 if (multi)
1181                         kfree(multi);
1182                 return -ENOENT;
1183         }
1184         if (ce->start > logical || ce->start + ce->size < logical) {
1185                 if (multi)
1186                         kfree(multi);
1187                 return -ENOENT;
1188         }
1189
1190         if (multi_ret) {
1191                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1192                                 GFP_NOFS);
1193                 if (!multi)
1194                         return -ENOMEM;
1195         }
1196         map = container_of(ce, struct map_lookup, ce);
1197         offset = logical - ce->start;
1198
1199         if (rw == WRITE) {
1200                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1201                                  BTRFS_BLOCK_GROUP_DUP)) {
1202                         stripes_required = map->num_stripes;
1203                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1204                         stripes_required = map->sub_stripes;
1205                 }
1206         }
1207         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1208             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1209                     /* RAID[56] write or recovery. Return all stripes */
1210                     stripes_required = map->num_stripes;
1211
1212                     /* Only allocate the map if we've already got a large enough multi_ret */
1213                     if (stripes_allocated >= stripes_required) {
1214                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1215                             if (!raid_map) {
1216                                     kfree(multi);
1217                                     return -ENOMEM;
1218                             }
1219                     }
1220         }
1221
1222         /* if our multi bio struct is too small, back off and try again */
1223         if (multi_ret && stripes_allocated < stripes_required) {
1224                 stripes_allocated = stripes_required;
1225                 kfree(multi);
1226                 multi = NULL;
1227                 goto again;
1228         }
1229         stripe_nr = offset;
1230         /*
1231          * stripe_nr counts the total number of stripes we have to stride
1232          * to get to this block
1233          */
1234         stripe_nr = stripe_nr / map->stripe_len;
1235
1236         stripe_offset = stripe_nr * map->stripe_len;
1237         BUG_ON(offset < stripe_offset);
1238
1239         /* stripe_offset is the offset of this block in its stripe*/
1240         stripe_offset = offset - stripe_offset;
1241
1242         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1243                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1244                          BTRFS_BLOCK_GROUP_RAID10 |
1245                          BTRFS_BLOCK_GROUP_DUP)) {
1246                 /* we limit the length of each bio to what fits in a stripe */
1247                 *length = min_t(u64, ce->size - offset,
1248                               map->stripe_len - stripe_offset);
1249         } else {
1250                 *length = ce->size - offset;
1251         }
1252
1253         if (!multi_ret)
1254                 goto out;
1255
1256         multi->num_stripes = 1;
1257         stripe_index = 0;
1258         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1259                 if (rw == WRITE)
1260                         multi->num_stripes = map->num_stripes;
1261                 else if (mirror_num)
1262                         stripe_index = mirror_num - 1;
1263                 else
1264                         stripe_index = stripe_nr % map->num_stripes;
1265         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1266                 int factor = map->num_stripes / map->sub_stripes;
1267
1268                 stripe_index = stripe_nr % factor;
1269                 stripe_index *= map->sub_stripes;
1270
1271                 if (rw == WRITE)
1272                         multi->num_stripes = map->sub_stripes;
1273                 else if (mirror_num)
1274                         stripe_index += mirror_num - 1;
1275
1276                 stripe_nr = stripe_nr / factor;
1277         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1278                 if (rw == WRITE)
1279                         multi->num_stripes = map->num_stripes;
1280                 else if (mirror_num)
1281                         stripe_index = mirror_num - 1;
1282         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1283                                 BTRFS_BLOCK_GROUP_RAID6)) {
1284
1285                 if (raid_map) {
1286                         int i, rot;
1287                         u64 tmp;
1288                         u64 raid56_full_stripe_start;
1289                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1290
1291                         /*
1292                          * align the start of our data stripe in the logical
1293                          * address space
1294                          */
1295                         raid56_full_stripe_start = offset / full_stripe_len;
1296                         raid56_full_stripe_start *= full_stripe_len;
1297
1298                         /* get the data stripe number */
1299                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1300                         stripe_nr = stripe_nr / nr_data_stripes(map);
1301
1302                         /* Work out the disk rotation on this stripe-set */
1303                         rot = stripe_nr % map->num_stripes;
1304
1305                         /* Fill in the logical address of each stripe */
1306                         tmp = stripe_nr * nr_data_stripes(map);
1307
1308                         for (i = 0; i < nr_data_stripes(map); i++)
1309                                 raid_map[(i+rot) % map->num_stripes] =
1310                                         ce->start + (tmp + i) * map->stripe_len;
1311
1312                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1313                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1314                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1315
1316                         *length = map->stripe_len;
1317                         stripe_index = 0;
1318                         stripe_offset = 0;
1319                         multi->num_stripes = map->num_stripes;
1320                 } else {
1321                         stripe_index = stripe_nr % nr_data_stripes(map);
1322                         stripe_nr = stripe_nr / nr_data_stripes(map);
1323
1324                         /*
1325                          * Mirror #0 or #1 means the original data block.
1326                          * Mirror #2 is RAID5 parity block.
1327                          * Mirror #3 is RAID6 Q block.
1328                          */
1329                         if (mirror_num > 1)
1330                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1331
1332                         /* We distribute the parity blocks across stripes */
1333                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1334                 }
1335         } else {
1336                 /*
1337                  * after this do_div call, stripe_nr is the number of stripes
1338                  * on this device we have to walk to find the data, and
1339                  * stripe_index is the number of our device in the stripe array
1340                  */
1341                 stripe_index = stripe_nr % map->num_stripes;
1342                 stripe_nr = stripe_nr / map->num_stripes;
1343         }
1344         BUG_ON(stripe_index >= map->num_stripes);
1345
1346         for (i = 0; i < multi->num_stripes; i++) {
1347                 multi->stripes[i].physical =
1348                         map->stripes[stripe_index].physical + stripe_offset +
1349                         stripe_nr * map->stripe_len;
1350                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1351                 stripe_index++;
1352         }
1353         *multi_ret = multi;
1354
1355         if (type)
1356                 *type = map->type;
1357
1358         if (raid_map) {
1359                 sort_parity_stripes(multi, raid_map);
1360                 *raid_map_ret = raid_map;
1361         }
1362 out:
1363         return 0;
1364 }
1365
1366 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1367                                        u8 *uuid, u8 *fsid)
1368 {
1369         struct btrfs_device *device;
1370         struct btrfs_fs_devices *cur_devices;
1371
1372         cur_devices = root->fs_info->fs_devices;
1373         while (cur_devices) {
1374                 if (!fsid ||
1375                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1376                         device = __find_device(&cur_devices->devices,
1377                                                devid, uuid);
1378                         if (device)
1379                                 return device;
1380                 }
1381                 cur_devices = cur_devices->seed;
1382         }
1383         return NULL;
1384 }
1385
1386 struct btrfs_device *
1387 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1388                            u64 devid, int instance)
1389 {
1390         struct list_head *head = &fs_devices->devices;
1391         struct btrfs_device *dev;
1392         int num_found = 0;
1393
1394         list_for_each_entry(dev, head, dev_list) {
1395                 if (dev->devid == devid && num_found++ == instance)
1396                         return dev;
1397         }
1398         return NULL;
1399 }
1400
1401 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1402                               struct btrfs_fs_devices *fs_devices)
1403 {
1404         struct map_lookup *map;
1405         u64 logical = BTRFS_SUPER_INFO_OFFSET;
1406         u64 length = BTRFS_SUPER_INFO_SIZE;
1407         int num_stripes = 0;
1408         int sub_stripes = 0;
1409         int ret;
1410         int i;
1411         struct list_head *cur;
1412
1413         list_for_each(cur, &fs_devices->devices) {
1414                 num_stripes++;
1415         }
1416         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1417         if (!map)
1418                 return -ENOMEM;
1419
1420         map->ce.start = logical;
1421         map->ce.size = length;
1422         map->num_stripes = num_stripes;
1423         map->sub_stripes = sub_stripes;
1424         map->io_width = length;
1425         map->io_align = length;
1426         map->sector_size = length;
1427         map->stripe_len = length;
1428         map->type = BTRFS_BLOCK_GROUP_RAID1;
1429
1430         i = 0;
1431         list_for_each(cur, &fs_devices->devices) {
1432                 struct btrfs_device *device = list_entry(cur,
1433                                                          struct btrfs_device,
1434                                                          dev_list);
1435                 map->stripes[i].physical = logical;
1436                 map->stripes[i].dev = device;
1437                 i++;
1438         }
1439         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1440         if (ret == -EEXIST) {
1441                 struct cache_extent *old;
1442                 struct map_lookup *old_map;
1443                 old = lookup_cache_extent(&map_tree->cache_tree,
1444                                           logical, length);
1445                 old_map = container_of(old, struct map_lookup, ce);
1446                 remove_cache_extent(&map_tree->cache_tree, old);
1447                 kfree(old_map);
1448                 ret = insert_cache_extent(&map_tree->cache_tree,
1449                                                    &map->ce);
1450         }
1451         BUG_ON(ret);
1452         return 0;
1453 }
1454
1455 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1456 {
1457         struct cache_extent *ce;
1458         struct map_lookup *map;
1459         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1460         int readonly = 0;
1461         int i;
1462
1463         ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1464         BUG_ON(!ce);
1465
1466         map = container_of(ce, struct map_lookup, ce);
1467         for (i = 0; i < map->num_stripes; i++) {
1468                 if (!map->stripes[i].dev->writeable) {
1469                         readonly = 1;
1470                         break;
1471                 }
1472         }
1473
1474         return readonly;
1475 }
1476
1477 static struct btrfs_device *fill_missing_device(u64 devid)
1478 {
1479         struct btrfs_device *device;
1480
1481         device = kzalloc(sizeof(*device), GFP_NOFS);
1482         device->devid = devid;
1483         device->fd = -1;
1484         return device;
1485 }
1486
1487 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1488                           struct extent_buffer *leaf,
1489                           struct btrfs_chunk *chunk)
1490 {
1491         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1492         struct map_lookup *map;
1493         struct cache_extent *ce;
1494         u64 logical;
1495         u64 length;
1496         u64 devid;
1497         u8 uuid[BTRFS_UUID_SIZE];
1498         int num_stripes;
1499         int ret;
1500         int i;
1501
1502         logical = key->offset;
1503         length = btrfs_chunk_length(leaf, chunk);
1504
1505         ce = search_cache_extent(&map_tree->cache_tree, logical);
1506
1507         /* already mapped? */
1508         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1509                 return 0;
1510         }
1511
1512         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1513         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1514         if (!map)
1515                 return -ENOMEM;
1516
1517         map->ce.start = logical;
1518         map->ce.size = length;
1519         map->num_stripes = num_stripes;
1520         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1521         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1522         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1523         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1524         map->type = btrfs_chunk_type(leaf, chunk);
1525         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1526
1527         for (i = 0; i < num_stripes; i++) {
1528                 map->stripes[i].physical =
1529                         btrfs_stripe_offset_nr(leaf, chunk, i);
1530                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1531                 read_extent_buffer(leaf, uuid, (unsigned long)
1532                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1533                                    BTRFS_UUID_SIZE);
1534                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1535                                                         NULL);
1536                 if (!map->stripes[i].dev) {
1537                         map->stripes[i].dev = fill_missing_device(devid);
1538                         printf("warning, device %llu is missing\n",
1539                                (unsigned long long)devid);
1540                 }
1541
1542         }
1543         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1544         BUG_ON(ret);
1545
1546         return 0;
1547 }
1548
1549 static int fill_device_from_item(struct extent_buffer *leaf,
1550                                  struct btrfs_dev_item *dev_item,
1551                                  struct btrfs_device *device)
1552 {
1553         unsigned long ptr;
1554
1555         device->devid = btrfs_device_id(leaf, dev_item);
1556         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1557         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1558         device->type = btrfs_device_type(leaf, dev_item);
1559         device->io_align = btrfs_device_io_align(leaf, dev_item);
1560         device->io_width = btrfs_device_io_width(leaf, dev_item);
1561         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1562
1563         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1564         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1565
1566         return 0;
1567 }
1568
1569 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1570 {
1571         struct btrfs_fs_devices *fs_devices;
1572         int ret;
1573
1574         fs_devices = root->fs_info->fs_devices->seed;
1575         while (fs_devices) {
1576                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1577                         ret = 0;
1578                         goto out;
1579                 }
1580                 fs_devices = fs_devices->seed;
1581         }
1582
1583         fs_devices = find_fsid(fsid);
1584         if (!fs_devices) {
1585                 ret = -ENOENT;
1586                 goto out;
1587         }
1588
1589         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1590         if (ret)
1591                 goto out;
1592
1593         fs_devices->seed = root->fs_info->fs_devices->seed;
1594         root->fs_info->fs_devices->seed = fs_devices;
1595 out:
1596         return ret;
1597 }
1598
1599 static int read_one_dev(struct btrfs_root *root,
1600                         struct extent_buffer *leaf,
1601                         struct btrfs_dev_item *dev_item)
1602 {
1603         struct btrfs_device *device;
1604         u64 devid;
1605         int ret = 0;
1606         u8 fs_uuid[BTRFS_UUID_SIZE];
1607         u8 dev_uuid[BTRFS_UUID_SIZE];
1608
1609         devid = btrfs_device_id(leaf, dev_item);
1610         read_extent_buffer(leaf, dev_uuid,
1611                            (unsigned long)btrfs_device_uuid(dev_item),
1612                            BTRFS_UUID_SIZE);
1613         read_extent_buffer(leaf, fs_uuid,
1614                            (unsigned long)btrfs_device_fsid(dev_item),
1615                            BTRFS_UUID_SIZE);
1616
1617         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1618                 ret = open_seed_devices(root, fs_uuid);
1619                 if (ret)
1620                         return ret;
1621         }
1622
1623         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1624         if (!device) {
1625                 printk("warning devid %llu not found already\n",
1626                         (unsigned long long)devid);
1627                 device = kzalloc(sizeof(*device), GFP_NOFS);
1628                 if (!device)
1629                         return -ENOMEM;
1630                 device->fd = -1;
1631                 list_add(&device->dev_list,
1632                          &root->fs_info->fs_devices->devices);
1633         }
1634
1635         fill_device_from_item(leaf, dev_item, device);
1636         device->dev_root = root->fs_info->dev_root;
1637         return ret;
1638 }
1639
1640 int btrfs_read_sys_array(struct btrfs_root *root)
1641 {
1642         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1643         struct extent_buffer *sb;
1644         struct btrfs_disk_key *disk_key;
1645         struct btrfs_chunk *chunk;
1646         struct btrfs_key key;
1647         u32 num_stripes;
1648         u32 len = 0;
1649         u8 *ptr;
1650         u8 *array_end;
1651         int ret = 0;
1652
1653         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1654                                           BTRFS_SUPER_INFO_SIZE);
1655         if (!sb)
1656                 return -ENOMEM;
1657         btrfs_set_buffer_uptodate(sb);
1658         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1659         array_end = ((u8 *)super_copy->sys_chunk_array) +
1660                     btrfs_super_sys_array_size(super_copy);
1661
1662         /*
1663          * we do this loop twice, once for the device items and
1664          * once for all of the chunks.  This way there are device
1665          * structs filled in for every chunk
1666          */
1667         ptr = super_copy->sys_chunk_array;
1668
1669         while (ptr < array_end) {
1670                 disk_key = (struct btrfs_disk_key *)ptr;
1671                 btrfs_disk_key_to_cpu(&key, disk_key);
1672
1673                 len = sizeof(*disk_key);
1674                 ptr += len;
1675
1676                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1677                         chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1678                         ret = read_one_chunk(root, &key, sb, chunk);
1679                         if (ret)
1680                                 break;
1681                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1682                         len = btrfs_chunk_item_size(num_stripes);
1683                 } else {
1684                         BUG();
1685                 }
1686                 ptr += len;
1687         }
1688         free_extent_buffer(sb);
1689         return ret;
1690 }
1691
1692 int btrfs_read_chunk_tree(struct btrfs_root *root)
1693 {
1694         struct btrfs_path *path;
1695         struct extent_buffer *leaf;
1696         struct btrfs_key key;
1697         struct btrfs_key found_key;
1698         int ret;
1699         int slot;
1700
1701         root = root->fs_info->chunk_root;
1702
1703         path = btrfs_alloc_path();
1704         if (!path)
1705                 return -ENOMEM;
1706
1707         /* first we search for all of the device items, and then we
1708          * read in all of the chunk items.  This way we can create chunk
1709          * mappings that reference all of the devices that are afound
1710          */
1711         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1712         key.offset = 0;
1713         key.type = 0;
1714 again:
1715         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1716         while(1) {
1717                 leaf = path->nodes[0];
1718                 slot = path->slots[0];
1719                 if (slot >= btrfs_header_nritems(leaf)) {
1720                         ret = btrfs_next_leaf(root, path);
1721                         if (ret == 0)
1722                                 continue;
1723                         if (ret < 0)
1724                                 goto error;
1725                         break;
1726                 }
1727                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1728                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1729                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1730                                 break;
1731                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1732                                 struct btrfs_dev_item *dev_item;
1733                                 dev_item = btrfs_item_ptr(leaf, slot,
1734                                                   struct btrfs_dev_item);
1735                                 ret = read_one_dev(root, leaf, dev_item);
1736                                 BUG_ON(ret);
1737                         }
1738                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1739                         struct btrfs_chunk *chunk;
1740                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1741                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1742                         BUG_ON(ret);
1743                 }
1744                 path->slots[0]++;
1745         }
1746         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1747                 key.objectid = 0;
1748                 btrfs_release_path(root, path);
1749                 goto again;
1750         }
1751
1752         ret = 0;
1753 error:
1754         btrfs_free_path(path);
1755         return ret;
1756 }
1757
1758 struct list_head *btrfs_scanned_uuids(void)
1759 {
1760         return &fs_uuids;
1761 }
1762
1763 static int rmw_eb(struct btrfs_fs_info *info,
1764                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1765 {
1766         int ret;
1767         unsigned long orig_off = 0;
1768         unsigned long dest_off = 0;
1769         unsigned long copy_len = eb->len;
1770
1771         ret = read_whole_eb(info, eb, 0);
1772         if (ret)
1773                 return ret;
1774
1775         if (eb->start + eb->len <= orig_eb->start ||
1776             eb->start >= orig_eb->start + orig_eb->len)
1777                 return 0;
1778         /*
1779          * | ----- orig_eb ------- |
1780          *         | ----- stripe -------  |
1781          *         | ----- orig_eb ------- |
1782          *              | ----- orig_eb ------- |
1783          */
1784         if (eb->start > orig_eb->start)
1785                 orig_off = eb->start - orig_eb->start;
1786         if (orig_eb->start > eb->start)
1787                 dest_off = orig_eb->start - eb->start;
1788
1789         if (copy_len > orig_eb->len - orig_off)
1790                 copy_len = orig_eb->len - orig_off;
1791         if (copy_len > eb->len - dest_off)
1792                 copy_len = eb->len - dest_off;
1793
1794         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1795         return 0;
1796 }
1797
1798 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1799                                 struct extent_buffer *orig_eb,
1800                                struct extent_buffer **ebs,
1801                                u64 stripe_len, u64 *raid_map,
1802                                int num_stripes)
1803 {
1804         struct extent_buffer *eb;
1805         u64 start = orig_eb->start;
1806         u64 this_eb_start;
1807         int i;
1808         int ret;
1809
1810         for (i = 0; i < num_stripes; i++) {
1811                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1812                         break;
1813
1814                 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1815                 if (!eb)
1816                         BUG();
1817                 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1818
1819                 eb->start = raid_map[i];
1820                 eb->len = stripe_len;
1821                 eb->refs = 1;
1822                 eb->flags = 0;
1823                 eb->fd = -1;
1824                 eb->dev_bytenr = (u64)-1;
1825
1826                 this_eb_start = raid_map[i];
1827
1828                 if (start > this_eb_start ||
1829                     start + orig_eb->len < this_eb_start + stripe_len) {
1830                         ret = rmw_eb(info, eb, orig_eb);
1831                         BUG_ON(ret);
1832                 } else {
1833                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1834                 }
1835                 ebs[i] = eb;
1836         }
1837 }
1838
1839 int write_raid56_with_parity(struct btrfs_fs_info *info,
1840                              struct extent_buffer *eb,
1841                              struct btrfs_multi_bio *multi,
1842                              u64 stripe_len, u64 *raid_map)
1843 {
1844         struct extent_buffer *ebs[multi->num_stripes], *p_eb = NULL, *q_eb = NULL;
1845         int i;
1846         int j;
1847         int ret;
1848         int alloc_size = eb->len;
1849
1850         if (stripe_len > alloc_size)
1851                 alloc_size = stripe_len;
1852
1853         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1854                             multi->num_stripes);
1855
1856         for (i = 0; i < multi->num_stripes; i++) {
1857                 struct extent_buffer *new_eb;
1858                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1859                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
1860                         ebs[i]->fd = multi->stripes[i].dev->fd;
1861                         multi->stripes[i].dev->total_ios++;
1862                         BUG_ON(ebs[i]->start != raid_map[i]);
1863                         continue;
1864                 }
1865                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1866                 BUG_ON(!new_eb);
1867                 new_eb->dev_bytenr = multi->stripes[i].physical;
1868                 new_eb->fd = multi->stripes[i].dev->fd;
1869                 multi->stripes[i].dev->total_ios++;
1870                 new_eb->len = stripe_len;
1871
1872                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1873                         p_eb = new_eb;
1874                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1875                         q_eb = new_eb;
1876         }
1877         if (q_eb) {
1878                 void *pointers[multi->num_stripes];
1879                 ebs[multi->num_stripes - 2] = p_eb;
1880                 ebs[multi->num_stripes - 1] = q_eb;
1881
1882                 for (i = 0; i < multi->num_stripes; i++)
1883                         pointers[i] = ebs[i]->data;
1884
1885                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1886         } else {
1887                 ebs[multi->num_stripes - 1] = p_eb;
1888                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1889                 for (j = 1; j < multi->num_stripes - 1; j++) {
1890                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1891                                 *(unsigned long *)(p_eb->data + i) ^=
1892                                         *(unsigned long *)(ebs[j]->data + i);
1893                         }
1894                 }
1895         }
1896
1897         for (i = 0; i < multi->num_stripes; i++) {
1898                 ret = write_extent_to_disk(ebs[i]);
1899                 BUG_ON(ret);
1900                 if (ebs[i] != eb)
1901                         kfree(ebs[i]);
1902         }
1903         return 0;
1904 }